Hello, On 2/27/2016 1:01 AM, RjOllos wrote: > > On Friday, February 26, 2016 at 2:18:15 PM UTC-8, Dirk Stöcker wrote: > > On Tue, 23 Feb 2016, Christian Boos wrote: > > > I've given some thoughts about maintaining a Genshi-compatibility > layer, > > but finally decided against it: > > Do I understand that correctly? You want to replace the template engine > with an incompatible new one and instead of operating these two in > parallel for at least one majore release, you want to drop the old one > immediatelly? > > > That will give a clear signal to the plugin developers, as well as > > giving them enough time to adapt before 1.4. > > Yes - it gives a clear signal: That you don't care about other > peoples work. > > > As usual your tone is quite unpleasant. People are more likely to care > about your opinion if you learn to phrase your points better.
Ah, glad to see it's not just me reading it that way ;-) Yes, please everyone, it would be great to keep a conversation tone where everyone feels welcome to share her opinion. Nothing has been "voted" upon yet, so I'm sure we can reach a consensus in what's the best outcome for the project, its maintainers, plugin developers and users. So yes, I want to switch to the Jinja2 template engine, this point seems already consensual, and it's been already the consensus 6 years ago in the first discussion on the topic (1). What has changed is that now we have a working branch which is going to be completed (2), and as things get more concrete we're also presented with a few development choices, with pros and cons. Dirk proposes to keep Genshi in parallel for "at least one major release" (1.3.x - 1.4.x). It's true that we did that when we switched from ClearSilver to Genshi. For that period, we had to keep telling people: "yeah, we know we have this and that issue with ClearSilver, please ask the plugin maintainer to switch to Genshi" and likewise with ClearSilver related bugs. Do we really want to do the same with Genshi again? Also, we would have to explain what still works and what no longer works, as 100% backward compatibility is not possible (ITemplateStreamFilter will no longer be supported). Having both template engines coexist is not impossible of course, as this is how things currently work in the branch, but this mixed situation has to be compared with the clarity of a clear-cut transition. Again, I'm not "viscerally" opposed to the idea, but I have to be convinced that it really makes a difference. I know that for plugins like the SpamFilter plugin, this would be "great", the Genshi templates would still work, no changes required, and the plugin can just be used unmodified. But what happens when 1.5.1 comes around? Same discussion? We're not going to maintain forever a template engine we don't use in core and know is problematic, so the necessity to migrate will come sooner or later. The question is, does it really make such a big difference if we announce it now, and have people do the migration early during the 1.3.x time frame, or if we say you'll have till 1.5.x do to it? > > Usually plugins are developed to solve a certain problem, not to > rewrite > the same code again and again when someone decides that a fancy new > library is again better than the last one. Genshi also was so much > better > than the previous one. > > > The migration to Jinja2 solves two real problems: > - Performance > - Genshi is no longer producing releases > > Additionally Jinja2 is much more widely adopted. It's not simply > "deciding a fancy new library is again better than the last one". You > should take a look at the performance numbers Christian has produced and > the detailed migration notes he has already prepared. This migration > solves several real problems and the amount of effort he has put into > documentation demonstrates that he clearly does care about helping users > migrate. > Yes, I think so far no one really spoke against the migration and we all see the benefits. I do what I can to prepare for a smooth transition. I even found yesterday that making the tag builder API replacement in trac.util.html compatible again with the Genshi templates was not as problematic as I initially thought. So now on the branch the remaining Genshi templates work again. I will continue to refactor trac.web.chrome to more cleanly separate the Genshi legacy support from the new code. An intermediate goal will be to make Genshi an *optional* package, i.e. Trac should be able to work fully without Genshi installed. > Compatibility is not a nuisance, it's required for acceptance. > > > Sometimes the ideal solution is not always practical. There will likely > be more than a year of pre-releases over which time plugins can be made > compatible. Maybe that's not a perfect solution, but a year is quite a > bit of time to plan your transition. I hope that after a few more exchanges we will be able to figure out the right approach, a clear severance between Trac and Genshi, or a middle-of-the-road approach, with Genshi templates still supported for a while in a "low profile" maintenance mode (i.e. if something's broken, tell people to migrate away from Genshi), and what's the appropriate time frame for either. But compatibility *will* be challenged, that's what happens with live software projects and I do hope we still count Trac as a live software project ;-) Genshi vs. Jinja2 is by far not the only compatibility concern for plugins in future versions. With Trac 1.3.x, I also support the proposal (made initially by Ryan? or me?) that we drop Python 2.6 compatibility and focus on Python 2.7. If we can merge in support for Python 3.5, all the better. We'll also pick at some point jQuery 3.0. And switch from XHTML to HTML5 (also more or less part of the jinja2 branch, btw). And yes, our data model still sucks and is eventually going to change radically one day. I can already here the "you're going to break *all* plugins with such a change!" criticisms. Well, yes, but we can keep that one for a future discussion ;-) -- Christian (1) - http://groups.google.com/group/trac-dev/msg/fc8d8c0447140110 (2) current output of python contrib/jinjastats.py: 58.54% converted to Jinja2 (5067 Jinja2 lines corresponding to 3861 Genshi lines on a total of 6596 Genshi lines) Yes, it looks like the way I write Jinja2 templates results in templates being 30% more verbose than their Genshi counterparts, but I think we also gain in clarity (and also I stick now to < 80 chars per line most of the time). -- You received this message because you are subscribed to the Google Groups "Trac Development" group. To unsubscribe from this group and stop receiving emails from it, send an email to trac-dev+unsubscr...@googlegroups.com. To post to this group, send email to trac-dev@googlegroups.com. Visit this group at https://groups.google.com/group/trac-dev. For more options, visit https://groups.google.com/d/optout.
